Method and device for providing hierarchical index of data structure of language model

FIELD: statistical language models, used in speech recognition systems.

SUBSTANCE: word indexes of bigrams are stored in form of common base with characteristic shifting. In one variant of realization, memory volume required for serial storage of bigram word indexes is compared to volume of memory, required for storage of indexes of bigram words in form of common base with characteristic shifting. Then indexes of bigram words are stored for minimization of size of data file of language model.

EFFECT: decreased memory volume needed for storing data structure of language model.

7 cl, 4 dwg

 

The technical field

This invention relates generally to the field of statistical language models used in speech recognition systems (CSR), in particular, relates to a method and device for more efficient organization of such models.

The level of technology

Known from the prior art system of continuous speech recognition work through the formation of a number of hypotheses sequence of words and calculate the probability of each sequence of words. Sequence with a low probability clipped, while sequences with a high probability of continuing. When decoding speech input is completed, the sequence with the highest probability is taken as the recognition result. Generally speaking, the origin is used, based on the probability. The reference sequence is the sum of the speaker relative to the amount of acoustic logarithms of the probabilities for all the minimal units of speech sounds or syllables) and linguistic reference (amounts linguistic logarithms of the probabilities for all words in the speech input).

In systems of continuous speech recognition typically use statistical n-g model language for developing statistical data. This model calculates the probability of n consecutive words in the specified area, pascalc is in practice we can assume, what's the current word depends on n preceding words. Anagramma model computes P(w), which is the probability for each word w. Bhrama model uses unigram and the conditional probability P(w2|w1), which is the probability of word w2provided that the preceding word is a word w1for each word w1and w2. Trigemina model uses unigram, bigram and the conditional probability P(w3|w2,w1), which is the probability of word w3provided that the previous two words are w1and w2for each word w3, w2and w1. Values Behrami and trigrammic probabilities calculated during the training of the language model that requires a huge amount of text data, a text body. The probability can be accurately estimated if the sequence of words appears relatively frequently in the training data. Such probabilities are conditionally existing. For n g probabilities that do not exist, use the recoil formula for the approximation of the value.

Such statistical language models are especially useful for recognition of continuous speech with a large vocabulary that recognize arbitrary speech (speech input). For example, theoretically words the OC, consisting of 50,000 words, you will be 50000 unigram, billions (500002) bigrams and more than 100 trillion (500003) trigrams. In practice, the number is significantly reduced, because bigram and trigrams exist only for word pairs and triples of words that appear relatively often. For example, in the English language for a well-known task, the wall Street Journal (WSJ) with a dictionary of 20,000 words in the language model is used only seven million bigrams and 14 million trigrams. These numbers depend on the specific language of the application and size of the text used for the development of language models. However, it is still a huge amount of data, and the size of the database language model and the method of data access has a significant impact on the vitality of speech recognition systems. Below is a description of a typical data structure of the language model with reference to figure 1.

Figure 1 shows trigemina data structure of the language model according to the prior art. Structure 100, shown in figure 1, contains origramy level 105, bigamy level 110 and trigeminy level 115. Write P(w3|w2,w1), where w3, w2and w1are the indexes of words that denotes the probability of word w3provided that the previous two words are the word of w1and the next word w2. ODA is dividing this probability of the word w 1located on anagrammer level 105, while origramy level contains a relationship with bigramma level. Get a pointer to the appropriate behramkale level 110 and determine bigram corresponding w1|w2while bigamy site contains a relationship with TriGranit level. Hence get a pointer to the appropriate trigrammaton level 115 and derive the probability P(w3|w2,w1) trigrams. Usually, unigram, bigram and trigrams data structures of the language model according to the prior art, all are stored in a simple sequential order and wanted consistently. Therefore, when searching for bigamy receive, for example, communication with bigramma level of anagramming level, and perform a sequential search Behrami to get the index for the second word.

The speech recognition systems are often carried out in small, compact computing systems such as personal computers, compact personal computers and even portable computer system. Such systems have limited capacity processing speed and volume data storage, so it is desirable to reduce the memory required to store the data structures of the language model.

Brief description of drawings

The present invention is described using an example, do not limit the him. The example described with reference to the accompanying drawings, in which similar elements are denoted by the same positions and in which is shown:

figure 1 - trigemina data structure of the language model according to the prior art;

figure 2 - block diagram of an example computing system 200 for implementing a database language models for continuous speech recognition, in accordance with this invention;

figure 3 - hierarchical storage structure according to a variant implementation of the present invention; and

4 is a graphical diagram of the process, according to a variant implementation of the invention.

Detailed description of embodiments of the invention

Below is a description of the improved data structure language model. The method according to this invention reduces the size of the file data of the language model. In one embodiment of the invention, the management information (for example, the index of the word) for behramkale level compress using a hierarchical begrammes storage structures. This invention is aimed at obtaining technical advantages from the fact that the distribution of word indexes for bigrams specific unigram are often within 255 indexes from each other (i.e., the offset may be represented by one byte). This allows you to store multiple word indexes as the basis of the s 2 bytes with an offset of one byte, as opposed to using three bytes to store each index word. Schemes, data compression, according to this invention is applicable, in particular, to behramkale level. This is because each unigram has on average approximately 300 bigrams, compared with approximately three trigrams for each bigram. That is, at the level of bigrams sufficient information for the practical implementation of hierarchical storage structure. In one embodiment, the hierarchy is used to store begrammes information only unigrams that are particularly relevant bigrams. Bigbamboo information for unigram with impractical small number of bigrams, is stored sequentially according to the prior art.

The method according to this invention can be extended to other applications of search-based indexes with a large number of indexes, where each index requires a large amount of memory.

Figure 2 shows a block diagram of an example computing system 200 for implementing a database language models for continuous speech recognition, in accordance with this invention. Different ways of calculating and comparing memory for storing data and a hierarchical structure of files, indexes, described in the estuaries and the e l e C can be made and used within the computing system 200, which may represent a public computer, a laptop computer or other similar device. Components of the computing system 200 is shown as an example, one or more components may be omitted or added. For example, you can use one or more memory devices for computer system 200.

As shown in figure 2, the computing system 200 includes a CPU 202 and a signal processor 203, connected to the display circuit 205, main memory 204, a static memory 206 and device 207 mass memory via the bus 201. Computing system 200 may also be connected to the display 221, a keyboard 222 input device 223 cursor control device 224 to obtain hard copies, devices 225 I/o and audio/speech device 226 via bus 201.

Bus 201 is a standard system bus for transferring data and signals. The CPU 202 and the signal processor 203 are processing units for computing system 200. The CPU 202 or the signal processor 203, or both can be used for processing data and/or signals to the computing system 200. The CPU 202 contains the block 231, the control unit 232 of arithmetic the tion logic and several registers 233, which are used for processing information and signals. The signal processor 203 may include components similar to the components of the Central processor 202.

The main unit memory 204 may be, for example, random access memory device (random-access memory, RAM) or other device dynamic memory for storing information or instructions (program code)that are used by the CPU 202 or the signal processor 203. The main unit memory 204 may store temporary variables or other intermediate information during execution of commands by the CPU 202 or the signal processor 203. Static memory 206 may be, for example, read-only memory (ROM) and/or other static storage device for storing information or commands that can be used by the CPU 202 or the signal processor 203. The device 207 mass storage may be, for example, a hard drive or floppy disk or optical disc drive for storing information or commands to the computing system 200.

The display 221 may be, for example, a cathode ray tube or liquid crystal display. The display 221 displays information or graphical information to the user. Computing Sistema can interact with the display 221 via interface display circuit 205. Keyboard 222 input is alphanumeric input device with an analog-to-digital Converter. Device 223 cursor control may be, for example, a mouse, a trackball, or cursor control keys to control the movement of the object on the display 221. The device 224 to obtain hard copies may be, for example, a laser printer for printing information on paper, film or other media. Computing system 200 may be connected to multiple devices 225 I/o. The hierarchical structure of index files of words according to this invention can be implemented with the help of hardware and/or software contained within computing system 200. For example, the Central processor 202 or signal processor 203 can execute codes or commands stored in a machine-readable storage medium, for example, the primary device 204 memory.

A machine-readable storage medium may contain a mechanism that provides (i.e. stores and/or transmits) information in machine readable form, such as a computer or digital processing device. For example, machine-readable media may include ROM, RAM, a means of storing data on magnetic disk storage medium Yes the data on the optical disk, the flash memory devices. Codes or commands can be represented by a signal carrier frequency, infrared signals, digital signals, and other similar signals.

Figure 3 shows the hierarchical structure data storage according to one variant of implementation of the present invention. The hierarchical structure 300 shown in figure 3, contains origramy level 310, bigamy level 320 and trigeminy level 330.

On anagrammer level 310, anagramma probability and return the weight - both are indices into a table of values and should not be subject to further reduction.

On average unigram have 300 bigrams that makes hierarchical storage practical, however, the individual unigram can have too little bigrams to justify working fields of the hierarchical structure. Unigram divided into two groups: unigram 311 with a sufficient number of bigrams in order hierarchical data storage bigrams was practical, and unigram 312 with too small number of matching bigrams in order hierarchical storage was practical. For example, for a task, the wall Street Journal (WSJ), with 19958 unigram, 16738 unigram have enough bigrams to justify hierarchical storage, and therefore bhrama information corresponding to this anagramma, is stored in a hierarchical order 321 bigrams. Such ungr the guide contain bigbamboo connection with hierarchical bigramma order 321. The rest 3220 unigram not have a sufficient number of bigrams to justify hierarchical storage, and therefore the corresponding bhrama information is stored in a simple sequential order. These unigram contain bigbamboo communication with serial bigramma order 322. In a typical text case there is very little unigram that do not have bigrams, and therefore they are not stored separately.

On bigmamma level 320 each bhrama (i.e. those that have corresponding trigrams) has a connection with TriGranit level 330. For typical text case there is relatively more bigrams that do not have trigrams than unigram that do not have bigrams. For example, in the wall Street Journal (WSJ), with 6850083 bigrams, 3414195 bigrams have corresponding trigrams, and 3435888 bigrams do not have a corresponding trigrams. In one embodiment, Behrami who do not have the trigrams are stored separately, eliminating four byte field relations of trigrams in these cases.

Typically, indexes words bigrams for one unigram are very close to each other. The proximity of these indexes is a typical feature of the language. This distribution of existing indexes bigrams allows you to split the index into groups, so that the displacement between the first index of word bigrams and last Indus is HMAC words bigrams is less than 256. Thus, this offset can be stored in one byte. It allows to represent, for example, the index words of three bytes as the sum of the bases of the two bytes and offset in one byte. Thus, since the two high order byte of the index words are repeated for several bigrams, these two bytes can be excluded from storage for some groups of bigrams. Such storage according to this invention provides significant compression on bigmamma level. As mentioned above, this does not apply to bigrams corresponding to each unigram. According to this invention calculates the amount of storage to determine whether it can be reduced by the hierarchical storage. If not, then the index of bigrams for specific unigram stored sequentially according to the prior art.

Figure 4 shows a graphical diagram of the process according to the variant of implementation of the present invention. The process 400 shown in figure 4, begins at operation 405 in which the estimate bigamy corresponding to a given unigram, to determine the amount of memory needed for a simple schematic sequential storage. In operation 410, the memory requirements for sequential storage compared to the memory requirements for storing hierarchical data structure. If there is no data compression (i.e. reduction of memory requirements), the Indus is si words bigrams remain consistently in operation 415. If a hierarchical data storage reduces the memory requirements, the indexes of words bigrams retain in the form of a common framework with a characteristic shift in operation 420. For example, for an index of words in three bytes, the total base may consist of two bytes with an offset of one byte.

The compression ratio depends on the probabilities of the bigrams in the language model. The language model used in the task, the wall Street Journal (WSJ), has approximately six million probabilities of bigrams that requires a memory capacity of approximately 97 MB. The implementation of hierarchical storage structure, according to this invention, provides compression to 32% indexes bigrams, which reduces the total amount of memory 12 MB (i.e. a total of about 11%). For other models of language, the compression ratio may be higher. For example, the implementation of hierarchical storage structure bigrams for language models for the task 863 Chinese language provides the compression index bigrams approximately 61.8%. This gives you the full compression ratio 26.7% (i.e. 70,3 MB compressed to 51.5 MB). This reduction of the data file language model significantly reduces the amount of memory and processing time.

Compression technology according to this invention is not practical to trigemina level, as there is, on average, only about three trigrams is but one bigram for language models for the task, the wall Street Journal (WSJ). Trigeminy level also contains no recoil weight or fields of communication, since there is no higher level.

This patent can be distributed for use in other scenarios, structured search, where the index word is the key; each index word requires a large amount of memory; and the number of index words is huge.

Although the invention is described with respect to several variants of implementation and illustrative figures, for specialists in the art will understand that the invention is not limited to the above variants of execution or specific shapes. In particular, the invention can be implemented in several alternative embodiments, which provide a hierarchical data structure to reduce the size of the database language model.

Therefore, it should be understood that the method and the device according to the invention can be implemented with various modifications and changes, within the attached claims. Thus, the present description should be considered as an illustration and not limitation of the invention.

1. The method of storing the set of indices of word bigrams in which each index word bigrams corresponds to a given unigram, in the form of a common framework with a characteristic offset, the index of words to bigram who are part trigeminal language model system for continuous speech recognition, and the language model simulates the task of the wall Street Journal, and the method comprises the following steps:

the mapping of the memory required for sequential storage of the set of indices of words, bigrams, and

the amount of memory required to store the hierarchical data structure of the set of indices of words, bigrams, and

the storing hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

2. The method according to claim 1, characterized in that the storage of hierarchical data structure of the set of indices of word bigrams includes storing each index word bigrams in the form of a common framework with a characteristic offset.

3. The method according to claim 2, characterized in that each index words bigram has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

4. A machine-readable storage medium on which is recorded commands, when the last processor performs a method of storing the set of indices of word bigrams, the index of word bigrams are part trigeminal language model system razpoznavam is I continuous speech, and the language model simulates the task of the wall Street Journal, and the method comprises the following steps:

the mapping of the memory required for sequential storage of the set of indices of words, bigrams, and

the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams; and

the storing hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

5. A machine-readable storage medium according to claim 4, characterized in that the storage of hierarchical data structure of the set of indices of word bigrams includes storing each index word bigrams in the form of a common framework with a characteristic offset.

6. A machine-readable storage medium according to claim 5, characterized in that each index words bigram has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

7. The device for storing the set of indices of word bigrams containing the processor and United with him the memory device, wherein the memory device stores commands, when executed by cataractarum the processor performs the mapping memory required for sequential storage of the set of indices of word bigrams, the index of word bigrams are part trigeminal language model system for continuous speech recognition, language model simulates the task of the wall Street Journal, and the amount of memory required to store the hierarchical data structure of the set of indices of words, bigrams, and provides storage hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

8. The device according to claim 7, characterized in that the storage of hierarchical data structure of the set of indices of word bigrams includes storing indexes of words bigrams corresponding to a given unigram, in the form of a common framework with a characteristic offset.

9. The device according to claim 8, characterized in that the index words of bigamy has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

10. The method of storing the set of indices of word bigrams consisting in that store the set of indices of word bigrams corresponding to a given unigram, in the form of a common framework with a characteristic offset, using the indexes of the words big the AMM, which are part trigeminal language model system for continuous speech recognition, the language model simulates the task 863 Chinese language.

11. The method according to claim 10, characterized in that each index words bigram has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

12. The method of storing the set of indices of word bigrams in which each index word bigrams corresponds to a given unigram, in the form of a common framework with a characteristic offset, and indexes words bigrams are part trigeminal language model system for continuous speech recognition, language model simulates the task 863 Chinese language, the method includes the following steps:

the mapping of the memory required for sequential storage of the set of indices of words, bigrams, and

the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams; and

the storing hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

13. The method according to item 12, wherein the storage s is aricescu data structure of the set of indices of word bigrams includes storing each index word bigrams in the form of a common framework with a characteristic offset.

14. The method according to item 13, wherein each index words bigram has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

15. A machine-readable storage medium on which a write command, when the last processor performs a method of storing the set of indices of word bigrams, the index of word bigrams are part trigeminal language model system for continuous speech recognition, language model simulates the task 863 Chinese language, and the method comprises the following steps:

the mapping of the memory required for sequential storage of the set of indices of words, bigrams, and

the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams; and

the storing hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

16. A machine-readable storage medium according to item 15, wherein storing the hierarchical data structure of the set of indices of word bigrams includes storing each index lovegra in the form of a common framework with a characteristic offset.

17. A machine-readable storage medium according to item 16, wherein each index words bigram has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.

18. The device for storing the set of indices of word bigrams, including the processor and United with him the memory device, wherein the memory device stores commands when the last processor performs the following actions: compares the amount of memory required for sequential storage of the set of indices of word bigrams, the index of word bigrams are part trigeminal language model for the recognition of continuous speech, and language model simulates the task 863 Chinese language, specifies the amount of memory required to store the hierarchical data structure of the set of indices of words, bigrams, and provides storage hierarchical data structure of the set of indices of word bigrams, if the amount of memory required to store the hierarchical data structure of the set of indices of word bigrams, was less than the amount of memory required for sequential storage of the set of indices of word bigrams.

19. The device according to p, characterized in that the storage of hierarchical data structure of the set of indices of word bigrams includes storing the Indus is xov words bigrams, corresponding to a given unigram, in the form of a common framework with a characteristic offset.

20. The device according to claim 19, characterized in that the index words of bigamy has a length of three bytes, the total Foundation has a length of two bytes and a typical offset has a length of one byte.



 

Same patents:

FIELD: speech activity transmission systems in distributed system of voice recognition.

SUBSTANCE: distributed system of voice recognition has voice recognition (VR) local mechanism in user unit and VR server mechanism in server. VR local mechanism has module for selection of features (FS), which selects features from voice signals. Voice activity detector (VAD) module detects voice activity invoice signal. Indication of voice activity is transmitted before features from user unit to server.

EFFECT: reduction in overloading of circuit; reduced delay and increased efficiency of voice recognition.

3 cl, 8 dwg, 2 tbl

FIELD: emotion detection.

SUBSTANCE: on the basis of input voice signal intensity, tempo and inflexion are recognized for every word in voice, values of changes are derived for the detected contents correspondingly and also signals which express the condition of each emotion of anger, sadness and satisfaction are derived on the basis of values of changes. Partner's emotion and information about the situation are introduced and thus information of instinctive motivation is generated. Besides, emotional information is generated which includes basic emotional parameters of satisfaction, anger and sadness, which is controlled on the basis of information about individual.

EFFECT: accurate detection of human emotion and the ability to generate sensitivity, close to human's sensitivity.

21 cl, 11 dwg

FIELD: computation engineering.

SUBSTANCE: device has object-oriented measurement tract correction mode. The device has test signal generator as sequence of N frequencies distributed over N strip means to which the hearing spectrum is divided with pauses between the frequencies, unit for transmitting test signals, microphone, N-bandwidth signal/noise ratio measurement unit, and computation unit for treating intelligibility. Manual generation frequency switch and pause switch mode is introduced. Level measurement unit is mounted in front of the transmitter. Controllable frequency characteristic adjustment unit is introduced in front of the signal/noise ratio measurement unit.

EFFECT: high accuracy and reliability of measurements.

2 dwg

FIELD: computation engineering.

SUBSTANCE: method involves carrying out object-oriented measurement tract correction. To carry out it, some test signal frequencies are transmitted of given level in turn. Real level of the acoustic signals is measured at each frequency at their transmission place. The acoustic signal levels are compared to given levels at the same frequencies, measured acoustic signals levels deviations from the given ones are measured and frequency characteristic of the measurement tract is adjusted to compensate the difference between the measured and given acoustic signal levels at their transmission place.

EFFECT: high accuracy and reliability of measurements.

2 dwg

FIELD: technology for measuring maximally possible legibility of speech in extremely unfavorable conditions, in noise, with low signal/noise ratios and is meant, primarily, for determining protection of objects in case of speech information loss through several leaking channels simultaneously.

SUBSTANCE: device contains generator of acoustic testing signal in form of series of N frequencies with pauses between frequencies, emitter, acoustic signal receiver, N-band device for measuring signal/noise rations and device for computing legibility. In addition, in parallel to receiver of acoustic signal K receivers of other signal types, formed by emission of acoustic testing signal, are mounted, for example, for detecting magnetic, electronic, vibro-acoustic types, while clock generator has additional outputs for organization of K+1 cycles of generator and synchronous control over switch of receivers and selection device.

EFFECT: increased trustworthiness of speech legibility measurements.

2 dwg

FIELD: technology for measuring maximally possible legibility of speech under extremely unfavorable conditions, in the noise, for low signal/noise ratios, and is meant mainly for determining protection of objects during loss of speech information through several leaking channels simultaneously.

SUBSTANCE: in accordance to method, in receipt point K receivers of other signal types, formed by acoustic testing signals, for example, electric, magnetic, vibro-acoustic signals, are positioned, acoustic and aforementioned K signal types formed in receipt position at all N frequencies of testing signal and in pauses between same, are received and measured separately, on basis of results of measurements signal/noise ratios are determined at each frequency of testing signal by each one of acoustic and K other types of received signals, at each frequency of testing signal greatest one of signal/noise relations are selected, measured by acoustic and K other types of received signals, then - speech legibility grade is calculated on basis of greatest signal/noise ratios.

EFFECT: increased trustworthiness of speech legibility measurements.

2 dwg

FIELD: technology for recognition of speech, in particular, method and device for dynamic adjustment of beam in Viterbi search.

SUBSTANCE: method includes selecting beginning width of beam, finding out whether value of probability for frame changes, dynamic adjustment of width of beam, decoding of input speech signal, while width of beam is dynamically adjusted. Also proposed is device including processor, component for speech recognition, connected to processor, memory, connected to processor. Component for speech recognition automatically adjusts width of beam to decode input speech signal.

EFFECT: increased speech recognition reliability.

6 cl, 6 dwg, 4 tbl

FIELD: technology for analyzing speech under unfavorable environmental conditions.

SUBSTANCE: during transformation of spoken command first circular buffer is continuously filled with digitized signal, comb of recursive filters is applied to multiply loosened signal and spectral components are utilized to fill second circular buffer, limits of speech fragment are determined within it on basis of adaptive estimate of noise environment, spectral components of speech fragment are transferred to linear analysis buffer, shortened sign space is received from aforementioned buffer and produced spectral components are compared to standard vectors of database commands.

EFFECT: utilization of device under conditions of, for example, moving vehicle or mechanical industry with high noise pollution level provides for stable recognition of commands independently on particularities of narrators pronunciation, decreased memory volume.

7 cl, 2 dwg

FIELD: engineering of informative system with speech interaction system.

SUBSTANCE: interaction system is connected to informative system and to recognition system. Interaction of user with informative system is performed by using user phrases, processed by recognition system. To achieve the result, system contains application sphere module, meant for setting phrase settings in it, which are constructed on basis of classes of objects, classes of attributes and classes of actions, which are common for aforementioned systems of subject application area and which should be matched by certain types of objects, types of actions, types of attributes and their instances, generated by informative system, for construction of grammar of phrases inputted by user.

EFFECT: possible interaction of user with informative system.

8 cl, 1 dwg

FIELD: analysis and recognition of speech signals, can be used for recognition of speech patterns.

SUBSTANCE: device for realization of aforementioned speech phoneme recognition method has: computing system, including clock generator, controller, random-access memory device, central microprocessor unit, meant for forming bispectral signs and recognizing them on basis of speech phonemes, digital-analog converter, long-term memorizing device, video-controller and analog-digital converter, and also keyboard, display, headphones and a microphone.

EFFECT: increased precision of speech patterns recognition due to forming of phoneme signs for speech phonemes recognition based on application of bispectral analysis, based on transformation of digital code series, appropriate for speech signals, to bispectral zone, characterizing interaction between values of Fourier components at different frequencies within speech spectrum, and thus, to provide selection of an additional, significantly new information from speech signals, to increase precision of phoneme recognition.

2 cl, 5 dwg

FIELD: communication systems; method for storing geographical information in communications center.

SUBSTANCE: geographical data is received, authentication query is sent to geographical data authentication database, which communicated with communications center. Answer for authentication query is received, and geographical data is stored in informational storage, which is a database, which communicates with communications center.

EFFECT: increased accuracy of service rendering corresponding to location in communication network on the basis of previously stored location information.

10 cl, 5 dwg

FIELD: information search means, database structures.

SUBSTANCE: two data areas are created. At least one of them is resident area, and at least one other area is non-resident for searched data object query source. Control data objects array is created in resident area, and/or control data objects array with corresponding to each object initial hyperlinks as linked data. In nonresident area control associated information data objects array is created and/or control associated information data objects array with corresponding to each object associated data and/or at least one secondary hyperlink.

EFFECT: simplified logical and physical database organization with permanent renewal of control associated information data objects, and increased performance of system due to simplified functioning of informational network communication nodes.

37 cl, 1 tbl

FIELD: computer engineering, automated system for collecting and processing electronic polls data.

SUBSTANCE: system consists of input messages receiving unit, data from server database receiving unit, election committee identification unit, first and second units for candidates base addresses identification, polls results disclosure time cycles selection unit, polls results recording time cycles selection unit, input messages receiving time cycles selection unit, database read and write signals forming unit, final polls results data forming unit.

EFFECT: increased system performance due to database entries address localization using receiving messages identifiers and forming of progressive total of polls results in real-time.

9 dwg

FIELD: computer engineering, systems for supporting informational identity of geographically distributed databases of airline companies.

SUBSTANCE: systems consists of address identifiers unit, memory area identification unit, input message target selection unit, database entries base address selection unit, adder, read signal forming unit, six registers, database entries identification unit, entries quantity identification unit, counter, control signal forming unit, OR elements.

EFFECT: increased system performance due to database entries addresses localization using data sources and flights identifiers.

9 dwg

FIELD: computer engineering; system for data distribution control in information analytical center network of air company commerce unit.

SUBSTANCE: system contains three registers, renewed data entries address identification device, client query data address identification device, decoder, data read control signal forming unit, and data output channels commutation unit.

EFFECT: simplified system, increased performance by excluding memory buffer blocks and time interval selector, and asynchronous mode of server-client interaction implementation.

5 dwg

FIELD: computer engineering; structure-statistical analysis of informational arrays.

SUBSTANCE: device contains current evaluation signal former, evaluation zones discriminator, pulse distributor, time intervals counter, commutator, search variable former, adding counters, storage units, division units, classifier, search strategy register, reset signals former, data representation and write unit, threshold signals change unit, current day timer, cutoff threshold former unit, structural analyzer.

EFFECT: increased informativity of traffic values, which define informational arrays structure due to structural analysis of determinate combinations.

2 cl, 3 dwg, 1 apl

FIELD: digital data processing using electrical devices; informational search means and database structures; foreign language learning.

SUBSTANCE: electronic documents data is entered into the computer. Electronic documents objects to be searched in informational system are discovered. Electronic documents objects to be searched in informational system are compared to informational system database objects. Electronic documents data is transformed by marking objects of electronic documents. Electronic documents with marked electronic documents data objects are visualized. Before entering electronic documents data into the computer data objects to be searched in informational system are entered, data objects to be searched in informational system are compared to informational system database data objects and buffer database of data objects is formed. Comparison of electronic documents objects to be searched in informational system is done with data objects in buffer database of data objects.

EFFECT: creation of effective method for searching and marking of information data, that is to implement the learning function "learned repetition", which is based on individual approach to learning.

18 cl

FIELD: distribution devices, terminal devices.

SUBSTANCE: in distribution device groups of two or more informational products which represent digital informational content are stored with information about policy administration which indicates user's rights to this group by interrelated method. Distribution device transfers the user requested informational content from group to the terminal device with license certificate (LC), refreshes information about policy administration decreasing policy validity. On return of the renewed LC distribution device increases the decreased policy validity taking into account the part of policy validity which is indicated in the renewed LC. On user's demand distribution device again transfers LC or other digital informational content.

EFFECT: distribution of digital content for a more complete satisfaction of user's demand.

22 cl, 58 dwg

FIELD: technology for setting up logical connections between a set of data files, such as files in Internet.

SUBSTANCE: device contains: display means; means for determining area of display for first file, made with possible selection of second area, which represent second files logically connected to first files, and for determining additional display files for other files, logically connected to first file; input means, made with possible selection of display area, appropriate for one of files. Procedures for assigning display areas may be performed iteratively, so that a certain diagram of interconnections between these areas is formed.

EFFECT: simplified setting up of connections between data files of various types.

6 cl, 34 dwg

FIELD: computer engineering, in particular, engineering of automated system for controlling fighting funds of national automatic system "Elections".

SUBSTANCE: system contains block for receiving input transactions, block for identification of addresses of fighting funds, block for modification of record addresses and reading data of fighting funds, block for identification of name parameters of deputy candidates, block for selection of type of financial operators, block for selection of upper limit of values of fighting funds, computing block, block for receiving data from server database, block for selecting lower limit of values of fighting funds, block for setting data dispensing modes, and block for integration of information signals.

EFFECT: increased speed of operation of system due to localization of addresses of documental data of fighting funds in database of system by identifiers of surname, name and patronymic name of deputy candidates and calculation of volumes of fighting funds directly in process of receipt of transaction data.

10 dwg

FIELD: data access technologies.

SUBSTANCE: method includes assignment of simplified network address, recording URL and converting numbers into storage system with net access, inputting assigned number into computer, transferring inputted number to storage system, converting number to URL, receiving page matching URL, and displaying it. Method for use in operation systems for message transfer include intercepting system level messages to certain objects and forming pseudonym messages during that. Systems realize said methods.

EFFECT: broader functional capabilities.

12 cl, 30 dwg

Up!